phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-07-31 08:00:09 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix prompt guard

Browse source
This commit is contained in:
Ashwin Bharambe 2024-10-03 09:17:11 -07:00
parent e9f6150588
commit ef591b44c8
2 changed files with 10 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

3
llama_stack/providers/impls/meta_reference/safety/config.py
View file

@ -47,8 +47,7 @@ class LlamaGuardShieldConfig(BaseModel):
return model return model
class PromptGuardShieldConfig(BaseModel): class PromptGuardShieldConfig(BaseModel): ...
model: str = "Prompt-Guard-86M"
class SafetyConfig(BaseModel): class SafetyConfig(BaseModel):

20
llama_stack/providers/impls/meta_reference/safety/safety.py
View file

@ -29,12 +29,16 @@ from .shields import (
ShieldBase, ShieldBase,
) )
PROMPT_GUARD_MODEL = "Prompt-Guard-86M"
def resolve_and_get_path(model_name: str) -> str: def resolve_and_get_path(model_name: str) -> str:
if model_name == PROMPT_GUARD_MODEL:
return model_local_dir(model_name)
model = resolve_model(model_name) model = resolve_model(model_name)
assert model is not None, f"Could not resolve model {model_name}" assert model is not None, f"Could not resolve model {model_name}"
model_dir = model_local_dir(model.descriptor()) return model_local_dir(model.descriptor())
return model_dir
class MetaReferenceSafetyImpl(Safety, RoutableProvider): class MetaReferenceSafetyImpl(Safety, RoutableProvider):
@ -45,7 +49,7 @@ class MetaReferenceSafetyImpl(Safety, RoutableProvider):
async def initialize(self) -> None: async def initialize(self) -> None:
shield_cfg = self.config.prompt_guard_shield shield_cfg = self.config.prompt_guard_shield
if shield_cfg is not None: if shield_cfg is not None:
model_dir = resolve_and_get_path(shield_cfg.model) model_dir = resolve_and_get_path(PROMPT_GUARD_MODEL)
_ = PromptGuardShield.instance(model_dir) _ = PromptGuardShield.instance(model_dir)
async def shutdown(self) -> None: async def shutdown(self) -> None:
@ -108,16 +112,10 @@ class MetaReferenceSafetyImpl(Safety, RoutableProvider):
disable_output_check=cfg.disable_output_check, disable_output_check=cfg.disable_output_check,
) )
elif typ == MetaReferenceShieldType.jailbreak_shield: elif typ == MetaReferenceShieldType.jailbreak_shield:
assert ( model_dir = resolve_and_get_path(PROMPT_GUARD_MODEL)
cfg.prompt_guard_shield is not None
), "Cannot use Jailbreak Shield since Prompt Guard not present in config"
model_dir = resolve_and_get_path(cfg.prompt_guard_shield.model)
return JailbreakShield.instance(model_dir) return JailbreakShield.instance(model_dir)
elif typ == MetaReferenceShieldType.injection_shield: elif typ == MetaReferenceShieldType.injection_shield:
assert ( model_dir = resolve_and_get_path(PROMPT_GUARD_MODEL)
cfg.prompt_guard_shield is not None
), "Cannot use PromptGuardShield since not present in config"
model_dir = resolve_and_get_path(cfg.prompt_guard_shield.model)
return InjectionShield.instance(model_dir) return InjectionShield.instance(model_dir)
elif typ == MetaReferenceShieldType.code_scanner_guard: elif typ == MetaReferenceShieldType.code_scanner_guard:
return CodeScannerShield.instance() return CodeScannerShield.instance()